1. Complete title of the Publication A Reinforcement Learning–Inspired Latent Yield-Based Adaptive Algorithm Switching Mechanism (Published at the International Conference on the Applications of Evolutionary Computation (Part of EvoStar 2026)) ------------------------------------------------------- 2. Author Information a) Jayprakash S. Nair, F-15, Indian Institute of Technology Guwahati, North Guwahati, Guwahati, Assam, India 781039 Phone:+918011223141 (jayprakash_24a02res06@iitp.ac.in) b) Jimson Mathew, Professor, Department of Computer Science and Engineering, Indian Institute of Technology (IIT) Patna, Bihta, Patna - 801103, Bihar, India Phone: +91-6115-233 347 (jimson@iitp.ac.in) c) Shivashankar B. Nair, Professor, Dept. of Computer Science and Engineering, Indian Institute of Technology Guwahati, North Guwahati, Guwahati, Assam, India 781039 Phone:+91-361-2582356 (sbnair@iitg.ac.in) ------------------------------------------------------- 3. Corresponding Author Jayprakash S. Nair (jayprakash_24a02res06@iitp.ac.in, Alternate: jsnair.hi@gmail.com) ------------------------------------------------------- 4. Paper Abstract Selecting the most suitable algorithm for a given problem instance remains a challenging task, particularly in online or dynamic environments where problem characteristics evolve over time. Relying solely on instantaneous performance metrics can result in a reactive and unstable behaviour, often leading to suboptimal algorithm switching. This paper introduces a computationally efficient approach for aggregating an algorithm’s performance across multiple problem instances that is fairly immune to erratic variations in instance features. Inspired by features inherent to Reinforcement Learning (RL), this technique encapsulates rewards and penalties into a latent yield that, in turn, triggers exploitation and exploration, consequently resulting in adaptive algorithm switching. The proposed technique employs island models, inspired by Genetic Algorithms, to facilitate parallel exploration and performance exchanges among algorithm populations inhabiting local repertoires. Experimental evaluations on sorting algorithms and robotic obstacle-avoidance tasks demonstrate the feasibility and effectiveness of the approach, highlighting its potential in domains where adaptive algorithm selection is critical. ------------------------------------------------------- 5. Competition Criterion (A) -------------------------------------------------------- 6. Statement: Why The Result Satisfies a) Criterion A - The result would qualify today as a patentable new invention Adaptive algorithm selection in dynamic online environments is a long-standing and inherently difficult problem in computer science and optimization. According to the No Free Lunch theorem, no single algorithm can consistently outperform all others across every possible problem domain or instance distribution. Consequently, selecting and adapting the most suitable algorithm during execution remains a major challenge, particularly in environments where problem characteristics evolve over time. The proposed framework introduces a novel adaptive algorithm-switching architecture combining Reinforcement Learning–inspired Latent-Yield accumulation with evolutionary Island-Model exploration. The system consists of multiple computational islands operating in parallel, where each island maintains: An agent that executes the algorithm, A repertoire of candidate algorithms, and A latent-yield storage or repository (Yielory) containing latent performance indicators (Yielons). The latent-yield mechanism enables a Wait–Watch–Switch strategy that reduces unstable or reactive switching behaviour typically caused by short-term performance fluctuations. Instead of relying solely on instantaneous rewards, the framework accumulates rewards and penalties into a latent yield over time that eventually guides the adaptive algorithm switching mechanism. Thus, unlike conventional rewards and penalties, this yield indirectly contains the memory of the overall performance of an algorithm. The latent-yield mechanism governs the balance between: Exploitation, where an algorithm continues execution due to sustained positive performance, and Exploration, where alternative algorithms are selected when performance deteriorates or stagnates. An agent within an island performs both intrinsic and extrinsic explorations, enabling it to explore algorithms from its own island’s repertoire as well as those belonging to other islands. The emergent adaptive switching behaviour enables the framework to autonomously manage algorithm selection without relying on manually engineered switching heuristics. A Central Interface Agent (CIA) coordinates interactions among the islands by ranking and monitoring island performance using latent-yield statistics. The framework was experimentally evaluated in two distinct domains: Adaptive selection among sorting algorithms operating on dynamically generated arrays, and Robotic obstacle-avoidance tasks within a simulated environment. Experimental results demonstrated the feasibility and effectiveness of the latent-yield-driven algorithm switching mechanism and the capability of the framework to maintain adaptive performance across heterogeneous environments. The work has been published in the proceedings of EvoApplications 2026, a peer-reviewed international conference in evolutionary computation. The integration of the novel latent-yield accumulation method, distributed island coordination, and the adaptive algorithm switching mechanism represents a potentially novel computational framework that could qualify as a patentable invention. -------------------------------------------------------- 7. Full Citation Nair, J.S., Mathew, J., Nair, S.B. (2026). A Reinforcement Learning–Inspired Latent Yield-Based Adaptive Algorithm Switching Mechanism. In: García-Sánchez, P., Díaz Álvarez, J., Murphy, A. (eds) Applications of Evolutionary Computation. EvoApplications 2026. Lecture Notes in Computer Science, vol 16524. Springer, Cham. https://doi.org/10.1007/978-3-032-23604-3_8 -------------------------------------------------------- 8. Prize Breakdown Statement The prize money, if any, is to be divided equally among the co-authors. -------------------------------------------------------- 9. Required Statement Indicating Why this Entry Could be the "Best". This entry addresses the long-standing and inherently difficult problem of adaptive algorithm selection in dynamic online environments. In accordance with the No Free Lunch theorem, no single algorithm performs optimally across all problem instances and environments. Consequently, designing systems capable of autonomously selecting and adapting algorithms over time remains a major challenge in optimization, artificial intelligence, and autonomous systems. The proposed framework introduces a novel latent-yield-driven adaptive switching mechanism inspired by Reinforcement Learning and Evolutionary Computation, constituting a parallel search for the best-suited algorithm. Unlike traditional switching approaches that rely heavily on instantaneous performance metrics, the proposed system accumulates rewards and penalties over time to create a stable latent-yield representation (Yielons) that guides both exploration and exploitation of algorithms within repertoires across multiple islands. A key distinguishing feature of the framework is the integration of: Parallel and Distributed Island-model exploration, Adaptive Latent-Yield accumulation, Autonomous algorithm switching, and Centralized inter-island coordination through the Central Interface Agent (CIA). The resulting architecture enables the system to dynamically manage multiple algorithms in parallel while reducing unstable knee-jerk switching behaviour common in reactive systems. The proposed framework was experimentally validated across heterogeneous domains, including adaptive sorting of random instances of arrays of numbers and robotic obstacle avoidance, demonstrating that the mechanism generalizes beyond a single application domain. The system autonomously selected and switched among candidate algorithms in response to changing environmental conditions without relying on manually engineered switching heuristics. Considering the novelty of the switching mechanism, accumulation of the latent yield, the inherent parallelism in the search, and the fact that it combines concepts from Reinforcement Learning, Evolutionary Computation, adaptive optimization, and Multi-Agent systems into a unified framework capable of autonomous decision-making in dynamic environments, this work could be categorized as among the best entries. Furthermore, the work has undergone independent peer review and has been published in the proceedings of the EvoApplications 2026 conference, providing external validation of its scientific contribution and novelty. -------------------------------------------------------- 10. Evolutionary Computation Type GA, Island Models -------------------------------------------------------- 11. Publication Date Appeared in Springer: 09 May 2026 Conference date: 8-10 April 2026 --------------------------------------------------------